kay retweeted
Since everyone is looking for a new agent.. 👀
🚀 Abacus Desktop 0.1.0 is out.
The abacus coding agent, now in a window: your projects and sessions, the agent's work as it streams, the diff, a terminal, and a browser it can use.
Open source. macOS, Windows, Linux. Any model.
abacus.empero.org
3
4
23
70,468
kay retweeted
This should tell you to stop using Hermes.
This is a compromised company.
This is a compromised company.
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
30
43
554
8,998
kay retweeted
我把nous research问的题目做成网页了
他们先洗脑模型，再出一些带有主观倾向恶意的题目要求模型回答再去打分，比如新疆，西藏，香港，广西洪水。事实率看他们的“事实”的就加分，不提或者相反的就0分/减分。宣传是看中国说法为主还是“事实”并列。总而言之，侬身上一股洋佬味
guileless-druid-a730a7.netlify.app
他们先洗脑模型，再出一些带有主观倾向恶意的题目要求模型回答再去打分，比如新疆，西藏，香港，广西洪水。事实率看他们的“事实”的就加分，不提或者相反的就0分/减分。宣传是看中国说法为主还是“事实”并列。总而言之，侬身上一股洋佬味
guileless-druid-a730a7.netlify.app
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
28
19
501
23,169
kay retweeted
Many questioned how well Manta performs under actual load.
So we gave it some: a real 7.4-million-token coding session, logged from a current frontier agent. 31,450 messages of training runs, logs and fixes.
We cut it, and let Manta write the next turn. ↓
So we gave it some: a real 7.4-million-token coding session, logged from a current frontier agent. 31,450 messages of training runs, logs and fixes.
We cut it, and let Manta write the next turn. ↓
8
16
79
292,570
kay retweeted
Ever wanted a model which can actually input large amounts of data and know what is in it? Manta will definitely be the model you need! For now it is only available for a certain group of selected people but who knows what the future will hold!👀
Introducing Manta, our long-context model.
Everything you give it, it keeps. Manta reads a hundred million tokens in one prompt.
Early access opens soon: empero.org/manta
🧵
0
1
5
140
Manta is coming.
It took a lot of work from a small team, and a lot of runs that didn't work before the ones that did. I'm proud of where it landed: a hundred million tokens in one prompt, and it still finds what you hid in there.
More to announce soon.
It took a lot of work from a small team, and a lot of runs that didn't work before the ones that did. I'm proud of where it landed: a hundred million tokens in one prompt, and it still finds what you hid in there.
More to announce soon.
Introducing Manta, our long-context model.
Everything you give it, it keeps. Manta reads a hundred million tokens in one prompt.
Early access opens soon: empero.org/manta
🧵
5
3
40
2,486
kay retweeted
Introducing Manta, our long-context model.
Everything you give it, it keeps. Manta reads a hundred million tokens in one prompt.
Early access opens soon: empero.org/manta
🧵
Everything you give it, it keeps. Manta reads a hundred million tokens in one prompt.
Early access opens soon: empero.org/manta
🧵
92
165
1,651
469,380
kay retweeted
self-hosted hf repo wasn't on my 2026 bingo card either
🤗 HuggingHub: your own self-hosted Hugging Face Hub in a single Rust binary.
Models, datasets, runnable Spaces, Git Smart HTTP + LFS, PRs & discussions, S3 or local disk 100% wire-compatible with huggingface_hub, transformers, and datasets.
GitHub: github.com/empero-org/hug…
0
1
14
904
kay retweeted
🤗 HuggingHub: your own self-hosted Hugging Face Hub in a single Rust binary.
Models, datasets, runnable Spaces, Git Smart HTTP + LFS, PRs & discussions, S3 or local disk 100% wire-compatible with huggingface_hub, transformers, and datasets.
GitHub: github.com/empero-org/hug…
Models, datasets, runnable Spaces, Git Smart HTTP + LFS, PRs & discussions, S3 or local disk 100% wire-compatible with huggingface_hub, transformers, and datasets.
GitHub: github.com/empero-org/hug…
5
9
36
3,116
i thought this was a joke. i really thought someone was trying to make fun of local models.
Modest intelligence is arriving faster than most people realize.
Today, we're open-sourcing TwIL-8B.
Just 8.8 billion parameters, yet approaching frontier-model scores on several difficult benchmarks:
• 93% of frontier performance on coding
• 92% on instruction following
matics
All from a model small enough to run on your own hardware.
But the benchmarks aren't the most important part of this announcement.
For years, the AI industry has focused on building the most capable model, regardless of the cost required to run it.
I believe that equation is changing.
The next era of AI will be defined by the cost of completing work. Not simply the intelligence of the model, but how efficiently that intelligence can accomplish the task.
If a model with 8 billion parameters can reliably complete the same work as one with hundreds of billions, why pay for the larger model?
As smaller models become more capable, intelligence becomes something we can distribute, specialize, and run across the hardware we already own.
That's the future we're building at @thewebAI
TwIL-8B is available today. Open weights.
And this is just the first of several announcements we're making today.
HF Link: huggingface.co/webAI-Official…
3
5
279
30,487
kay retweeted
This is actually so fun to play around with! Im so excited for the next releases wish i could reveal more
Meet Pi and Vo, the two voices of PiVo Studio.
Local text-to-speech in a single Rust binary: ~4× faster than real time on one CPU thread, first audio in ~30 ms, OpenAI-compatible API. Runs on desktop, iOS, Android and in the browser.
github.com/empero-org/PiV…
1
1
10
347